Image Classification on STL-10 (Top-1 and Top-5 Accuracy)
82.89Top-1 AccSimCLR
Evaluation results
| Method | Links | ||
|---|---|---|---|
| SimCLREvaluation Protocol=Linear probe, Backbone=ResNet-18, Epochs=200, Batch Size=1282026.01 | 82.89 | 99.36 | |
| BTEvaluation Protocol=Linear probe, Backbone=ResNet-182026.01 | 82.19 | 99.04 | |
| HypersolidEvaluation Protocol=Linear probe, Backbone=ResNet-182026.01 | 82.11 | 99.2 | |
| VICRegEvaluation Protocol=Linear probe, Backbone=ResNet-182026.01 | 81.69 | 99.14 | |
| BYOLEvaluation Protocol=Linear probe, Backbone=ResNet-182026.01 | 81.28 | 99.28 | |
| LeJEPAEvaluation Protocol=Linear probe, Backbone=ResNet-182026.01 | 80.73 | 99.33 | |
| DINOEvaluation Protocol=Linear probe, Backbone=ResNet-182026.01 | 79.75 | 99.11 | |
| SimCLREvaluation Protocol=k-NN, K=5, Backbone=ResNet-18, Epochs=200, Batch Size=1282026.01 | 79.16 | 94.91 | |
| VICRegEvaluation Protocol=k-NN, K=5, Backbone=ResNet-182026.01 | 78.48 | 93.85 | |
| LeJEPAEvaluation Protocol=k-NN, K=5, Backbone=ResNet-182026.01 | 78.28 | 94.78 | |
| BTEvaluation Protocol=k-NN, K=5, Backbone=ResNet-182026.01 | 78.2 | 93.98 | |
| HypersolidEvaluation Protocol=k-NN, K=5, Backbone=ResNet-182026.01 | 77.6 | 93.58 | |
| BYOLEvaluation Protocol=k-NN, K=5, Backbone=ResNet-182026.01 | 77.14 | 94.39 | |
| DINOEvaluation Protocol=k-NN, K=5, Backbone=ResNet-182026.01 | 76.85 | 93.53 | |
| ThermoLion2025.12 | 74.22 | — | |
| SUPERVISEDEvaluation Protocol=k-NN, K=5, Backbone=ResNet-182026.01 | 72.43 | 87.69 | |
| SUPERVISEDEvaluation Protocol=Linear probe, Backbone=ResNet-182026.01 | 71.86 | 97.41 | |
| SWATS2025.12 | 56.04 | — | |
| Adam (Baseline)2025.12 | 55.94 | — | |
| AdamW2025.12 | 55.34 | — | |
| Lookahead2025.12 | 50.6 | — | |
| MuAdam2025.12 | 50.52 | — | |
| RMSprop2025.12 | 47.38 | — | |
| Lion2025.12 | 46.54 | — |